Handle thread pool queue rejection better

A few set of tests fail due to not having a queue large enough for the cpu thread pool, which in turn runs the command on the persistence thread pool. We don't want this to happen as this can cause a deadlock.

https://ci.infinispan.org/job/Infinispan/job/master/1567/testReport/junit/org.infinispan.persistence/ClusteredConditionalCommandTest/testPutIfAbsentOnPrimaryOwnerWithIgnoreReturnValues_DIST_SYNC__tx_false_/